01·Code: Copilot-style and repo context

Introduction

Case study: how Copilot, Cursor, Aider, Claude Code and agentic IDEs build prompts for code generation. FIM, neighboring tabs, repo-aware retrieval, multi-file edits. Code-specific failure modes: API hallucination, context decay, subtle wrong code. Test-driven prompting, type-aware completion, security-aware generation. SWE-bench vs HumanEval. Project rules files (.cursorrules/CLAUDE.md/AGENTS.md). Production cost optimisation.